[jbosscache-dev] Re: Helping out with JBCACHE-991 (Non-tx batching)

[Manik Surtani]

On 1 Aug 2007, at 23:00, Vincent Marquez wrote:

> >I'd have thought batching would hold locks as well.
> >
> >Unless you see batching as just queueing calls and applying them when
> >batchComplete() (or whatever we call it) is invoked.  But then how do
> >you deal with failures?  Is the batch atomic?
>
> Well, I was assuming they wouldn't hold locks (at least until the  
> commit occured).  I figured
> a batch might be long running with adds, and with the current  
> locking scheme, other reads would block
> until the commit for the nodes that were added/updated, no?

Yes, that's correct, unless you use optimistic locking.

>
> You could have the commit be optimistic with atomic behavior, or  
> non atomic and just have exception notifications.  Are either of  
> these options?

At the moment optimistic locking is atomic, so even if there is a  
version mismatch for one operation, the entire transaction (i.e.,  
batch) will fail.

Just like with databases, I'm assuming batch operations would only be  
write operations and not read operations?


--
Manik Surtani

Lead, JBoss Cache
JBoss, a division of Red Hat